I've been working on a Teradata site now for the past 2 years and continue to have differing views on the use of the TD application. I'm after some advice from the wider community on the value of statistics.
When confronted with a view that statistics have little value, that we should write our SQL so we don't need to rely on statistics, that our SQL should only contain 1 join and the result loaded to a physical work table and then subsequently updated by other single join SQL statements.
To me this view is dumbing the use of the teradata application to it's lowest level and not using it as it was intended. I'm interested to know how others would respond to being forced down a similar path.
Regarding publishing of the first post:
When it will be published depends on when a moderator logged in and marked your post as "no spam". Newly registered users need to be approved as 99+% of all new accounts are spammers :-( After this apporval you can post at will.
Regarding your original question:
You mean your supposed to split multiple joins into binary relations and materialize each result of a join into a temporary table (hopefully a Volatile Table)?
Ouch. Your absolutely right, this is just dumb.
One of the big advantages of RDBMSes is that you don't need to find out how data is extracted in the most efficient way, there's an optimizer for it. And all optimizers are cost-based nowadays, but cost calculation without knowing costs (i.e. no information about number of rows returned by a WHERE-condition, etc.) is really hard to do.
Btw, this approach would be similar stupid on any existing RDBMS :-)
Appreciate your feedback and very much in line with my views and experience using DB2 for 10 years.
We are wasting so much time and effort negotiating away from this type of usage, which to me is just crazy. With so much written about statistics and management methods it just doesn't make sense.